“In a famous talk associated with the sequence-to-sequence paper, Ilya Sutskever claimed that given enough supervised data and a large neural network, one could solve almost all problems.”